extract kei substr group featur text classif dell zhang wee sun lee text classif applic appeal document string charact bag word previou research studi area focus variant gener markov chain model discrimin machin learn method support vector machin svm success text classif word featur effect effici appli straightforwardli take substr corpu featur paper propos partit substr statist equival group pick group import statist sens featur name kei substr group featur text classif propos suffix tree base algorithm extract featur linear time respect total number charact corpu experi english chines greek dataset show svm kei substr group featur achiev outstand perform text classif task 